Yutaka Akiyama Chief, TRC Massively Parallel Application Laboratory

Profile

Affiliation:	Massively Parallel Application Laboratory, Massively Parallel Systems Department, Tsukuba Research Center, RWCP
Date of birth:	May 30, 1961
Birthplace:	Tokyo
Background:
1984	Graduated from the Electric Engineering Department, Faculty of Engineering, Keio University
1990	Completed the Doctorate Course of Electric Engineering, Postgraduate School of Engineering, Keio University; Ph. D.
1990	Joined the Electrotechnical Laboratory, Agency of Industrial Science and Technology, Ministry of International Trade and Industry, and assigned to the Computation Models Section
1992	Appointed to associate professor of Institute for Chemical Research, Kyoto University
1996	Joined the Real World Computing Partnership

I joined RWCP's Tsukuba Research Center last April. In the Massively Parallel Application Laboratory, we are developing large-scale application programs to demonstrate parallel computing technology. Since I have been involved in information analysis of genes and proteins in my career, I have chosen many biology-related subjects in this laboratory. In addition, we plan to work on other themes including large-scale problems in electro-magnetic field simulations and combinational optimization problems in industrial applications.

Our overall goal is to implement practical programs running on parallel computers to be used by experts in specific areas. To achieve this goal, we are gathering skilled members with various backgrounds. In particular, we need participants from private companies for the development of industrial application programs.

Until last March, I was engaged in the Human Genome Project of the Ministry of Education as a member from Kyoto university. A tremendous amount of calculation and data processing power and resources are required to combine, parse, compare with past examples and interpret data on genetic information (DNA sequence) which increases exponentially. State-of-the-art information technology and power of parallel computers are indispensable for the purpose.

The desperate need of researchers in biology for more computer power made me believe that the current computer engineering was far from mature. On the other hand, I found that researchers in some application areas were often short-sighted and their approach was a roundabout way towards solving problems. I learned both the joys and difficulties of interdisciplinary effort through this project. Though I belonged to the biology-related institute at that time, I never touched a test tube but was working among supercomputers and parallel computers.

Upon completion of the project, I decided to return from the application side to the computer side as a researcher of computer science. It was a drastic decision for me and my family to resign as associate professor of the university and obtain a position in the RWCP. It was not an impulsive decision because I had been very much interested in joining the RWC project long before. The interest was probably instilled in me because I have seen that many researchers around me actively engaged in the national projects.

My first encounter with computers took place when I was a junior high school student. A UNIVAC 1106 and TOSBAC 3400 of the university's computer center became my closest partners, perhaps this was even first love. Although their memory capacities were very limited, they seemed almighty to me at that time. However, they were suddenly turned out to be useless when given problems where the number of possible combinations explosively increased. My small allowance was used up in computer fees.

When I entered Keio University, I started to research on parallel algorithm for VLSI such as systolic array under the guidance of Professor Aiso, and was optimistically attracted by fine-grain parallelism. During the doctorate course, I finally started studying neural networks - the subject I had been interested in for years. Under the guidance of Professor Aiso and Professor Anzai, I proposed a computation model named a "Gaussian machine" which combined a Boltzmann machine and Hopfield network, and I applied the theory on approximate solutions of the optimization problem. My approach was to minimize the objective function of the optimization problem, which was essentially the matter of "0 or 1," by replacing it with continuous functions of which the extremes were almost the same as those of the objective function. Smoother functions should be used during the initial stages of calculation and then functions less smooth but closer to the original objective function should be gradually used in later stages. This gives the optimum and/or quasi-optimum solutions with surprisingly high probabilities.
As for the simple N-queen problem, for example, the solution can easily be obtained even when N is in the order of thousands. This approach is effective for problems that contain a large number of hypotheses in both a competitive and cooperative manner. With its parallel nature of various levels, it is especially suitable for parallel computers.

I started applying this approach to genetic information analyses when I was working for the Electrotechnical Laboratory because I thought the nature of the problem space was straightforward and a deep basin existed around its optimum solution.

In an effort to get slightly closer to the prediction of the three-dimensional structure of proteins, which is one of the hardest problems supposedly worthy of the Nobel Prize, I shall try to combine the parallelization of biophysical calculations with such powerful approximation approaches in the RWC project.